[X86] Add PTEST tests showing failure to extract allsign cases

As discussed on PR42035, we can often use MOVMSK to avoid a cmpgt/ashr by just analysing the extracted signbits.

diff --git a/llvm/test/CodeGen/X86/combine-ptest.ll b/llvm/test/CodeGen/X86/combine-ptest.ll
index d23277f..2928023 100644 (file)
--- a/llvm/test/CodeGen/X86/combine-ptest.ll
+++ b/llvm/test/CodeGen/X86/combine-ptest.ll
@@ -1,5 +1,6 @@
 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx | FileCheck %s
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx  | FileCheck %s --check-prefixes=CHECK,AVX1
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2 | FileCheck %s --check-prefixes=CHECK,AVX2
 
 ;
 ; testz(~X,Y) -> testc(X,Y)
@@ -297,6 +298,102 @@ start:
   ret i1 %6
 }
 
+;
+; TODO: testz(ashr(X,bw-1),-1) -> movmsk(X)
+;
+
+define i32 @ptestz_v2i64_signbits(<2 x i64> %c, i32 %a, i32 %b) {
+; CHECK-LABEL: ptestz_v2i64_signbits:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    movl %edi, %eax
+; CHECK-NEXT:    vpxor %xmm1, %xmm1, %xmm1
+; CHECK-NEXT:    vpcmpgtq %xmm0, %xmm1, %xmm0
+; CHECK-NEXT:    vptest %xmm0, %xmm0
+; CHECK-NEXT:    cmovnel %esi, %eax
+; CHECK-NEXT:    retq
+  %t1 = ashr <2 x i64> %c, <i64 63, i64 63>
+  %t2 = call i32 @llvm.x86.sse41.ptestz(<2 x i64> %t1, <2 x i64> <i64 -1, i64 -1>)
+  %t3 = icmp ne i32 %t2, 0
+  %t4 = select i1 %t3, i32 %a, i32 %b
+  ret i32 %t4
+}
+
+define i32 @ptestz_v8i32_signbits(<8 x i32> %c, i32 %a, i32 %b) {
+; AVX1-LABEL: ptestz_v8i32_signbits:
+; AVX1:       # %bb.0:
+; AVX1-NEXT:    movl %edi, %eax
+; AVX1-NEXT:    vpsrad $31, %xmm0, %xmm1
+; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm0
+; AVX1-NEXT:    vpsrad $31, %xmm0, %xmm0
+; AVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm1, %ymm0
+; AVX1-NEXT:    vptest %ymm0, %ymm0
+; AVX1-NEXT:    cmovnel %esi, %eax
+; AVX1-NEXT:    vzeroupper
+; AVX1-NEXT:    retq
+;
+; AVX2-LABEL: ptestz_v8i32_signbits:
+; AVX2:       # %bb.0:
+; AVX2-NEXT:    movl %edi, %eax
+; AVX2-NEXT:    vpsrad $31, %ymm0, %ymm0
+; AVX2-NEXT:    vptest %ymm0, %ymm0
+; AVX2-NEXT:    cmovnel %esi, %eax
+; AVX2-NEXT:    vzeroupper
+; AVX2-NEXT:    retq
+  %t1 = ashr <8 x i32> %c, <i32 31, i32 31, i32 31, i32 31, i32 31, i32 31, i32 31, i32 31>
+  %t2 = bitcast <8 x i32> %t1 to <4 x i64>
+  %t3 = call i32 @llvm.x86.avx.ptestz.256(<4 x i64> %t2, <4 x i64> <i64 -1, i64 -1, i64 -1, i64 -1>)
+  %t4 = icmp ne i32 %t3, 0
+  %t5 = select i1 %t4, i32 %a, i32 %b
+  ret i32 %t5
+}
+
+define i32 @ptestz_v8i16_signbits(<8 x i16> %c, i32 %a, i32 %b) {
+; CHECK-LABEL: ptestz_v8i16_signbits:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    movl %edi, %eax
+; CHECK-NEXT:    vpsraw $15, %xmm0, %xmm0
+; CHECK-NEXT:    vptest %xmm0, %xmm0
+; CHECK-NEXT:    cmovnel %esi, %eax
+; CHECK-NEXT:    retq
+  %t1 = ashr <8 x i16> %c, <i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15>
+  %t2 = bitcast <8 x i16> %t1 to <2 x i64>
+  %t3 = call i32 @llvm.x86.sse41.ptestz(<2 x i64> %t2, <2 x i64> <i64 -1, i64 -1>)
+  %t4 = icmp ne i32 %t3, 0
+  %t5 = select i1 %t4, i32 %a, i32 %b
+  ret i32 %t5
+}
+
+define i32 @ptestz_v32i8_signbits(<32 x i8> %c, i32 %a, i32 %b) {
+; AVX1-LABEL: ptestz_v32i8_signbits:
+; AVX1:       # %bb.0:
+; AVX1-NEXT:    movl %edi, %eax
+; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
+; AVX1-NEXT:    vpxor %xmm2, %xmm2, %xmm2
+; AVX1-NEXT:    vpcmpgtb %xmm1, %xmm2, %xmm1
+; AVX1-NEXT:    vpcmpgtb %xmm0, %xmm2, %xmm0
+; AVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
+; AVX1-NEXT:    vptest %ymm0, %ymm0
+; AVX1-NEXT:    cmovnel %esi, %eax
+; AVX1-NEXT:    vzeroupper
+; AVX1-NEXT:    retq
+;
+; AVX2-LABEL: ptestz_v32i8_signbits:
+; AVX2:       # %bb.0:
+; AVX2-NEXT:    movl %edi, %eax
+; AVX2-NEXT:    vpxor %xmm1, %xmm1, %xmm1
+; AVX2-NEXT:    vpcmpgtb %ymm0, %ymm1, %ymm0
+; AVX2-NEXT:    vptest %ymm0, %ymm0
+; AVX2-NEXT:    cmovnel %esi, %eax
+; AVX2-NEXT:    vzeroupper
+; AVX2-NEXT:    retq
+  %t1 = ashr <32 x i8> %c, <i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7>
+  %t2 = bitcast <32 x i8> %t1 to <4 x i64>
+  %t3 = call i32 @llvm.x86.avx.ptestz.256(<4 x i64> %t2, <4 x i64> <i64 -1, i64 -1, i64 -1, i64 -1>)
+  %t4 = icmp ne i32 %t3, 0
+  %t5 = select i1 %t4, i32 %a, i32 %b
+  ret i32 %t5
+}
+
 declare i32 @llvm.x86.sse41.ptestz(<2 x i64>, <2 x i64>) nounwind readnone
 declare i32 @llvm.x86.sse41.ptestc(<2 x i64>, <2 x i64>) nounwind readnone
 declare i32 @llvm.x86.sse41.ptestnzc(<2 x i64>, <2 x i64>) nounwind readnone